Perform benchmarking on in silico mixture results.
Contains code to regenerate figures 2b-d, 3, 4 and supplementary figures 8-9 from the SVclone paper.
Optimal versus ground truth CCF distributions. From left to right, by column: i) SV theoretical mixture truth CCFs, ii) ‘optimal’ SV CCFs based on sample membership, iii) SNV theoretical mixture truth CCF, iv) ‘optimal’ SNV CCFs based on sample membership.
This figure is an extended version of the figure that appears in the paper.
CCF mean error
## data V1
## 1: SVs 0.092590175
## 2: SNVs 0.002067377
True minus optimal CCF averages for SNVs (red) and SVs (blue).
ROC curves showing sensitivity and specificity of classifying variants as subclonal:
Respective AUCs
## SV_ccf_mean SV_ccf_min SV_ccf_max SNV_ccf
## 0.9083199 0.8862681 0.9161043 0.9039557
SV CCF mean optimal cutoff
## [1] 0.6910936
SV CCF min optimal cutoff
## [1] 0.6515464
SV CCF min optimal cutoff
## [1] 0.7796553
SNV CCF optimal cutoff
## [1] 0.7182253
Percent of SVs with heterogenous copy-number states across both SV sides.
## mix heterogenous_CN_percent
## 1: 001bM 0.31
## 2: 001gM 0.38
## 3: 0.1-0.9 0.26
## 4: 0.2-0.8 0.25
## 5: 0.3-0.7 0.27
## 6: 0.4-0.6 0.25
## 7: 0.5-0.5 0.25
## 8: 0.6-0.4 0.24
## 9: 0.7-0.3 0.24
## 10: 0.8-0.2 0.25
## 11: 0.9-0.1 0.25
Performance comparison of the dual-end SV model versus single-end.
Comparing ROC curves of dual-end and single-end models.
Respective AUCs
## SVs_dual_end SVs_single_end
## 0.8233912 0.8095298
Performance for SVs that have clonal background copy-numbers vs. all SVs (clonal + subclonal background copy-numbers).
Performance across SVclone (SVs and SNVs), battenberg and PyClone.
Performance of post-assignment using a joint SV + SNV model, compared with SV and SNV results wihout post-assign. PyClone results added for reference.
PyClone precision traces.
Comparison of clustering metrics in the unmodified dual-end model under various background copy-number perturbations (+1 and -1 from major allele copy-number, and +0.3 or -0.3 from the copy-number fraction (for subclonal copy-numbers).
Comparison of clustering metrics for the dual and single-end models under copy-number perturbation.